基于堆叠双向LSTM的雷达目标识别方法

Radar Target Recognition Based on Stacked Bidirectional LSTM

摘要:

现阶段雷达目标检测识别主要依赖人工算法提取目标的特征，难点在于环境自适应能力弱，高强度杂波背景下难以有效检测到目标。针对上述问题，结合深度学习在图像识别等领域表现出的强大的学习表示能力，提出基于堆叠双向长短期记忆网络（Long Short-Term Memory Network，LSTM）的雷达目标识别方法。网络模型以雷达多普勒维的回波数据构建数据集，采用双向LSTM提取雷达回波数据在时间序列上的正向和逆向信息，通过RMSProp优化算法对神经网络参数迭代训练，实现了对无人机这种低空慢速小目标的有效识别。实验结果表明，基于堆叠双向LSTM的雷达目标识别方法优于传统的SVM分类算法和卷积神经网络分类算法。

Abstract:

At this stage, radar target detection and recognition mainly rely on artificial algorithms to extract the target's characteristics. The difficulty lies in the weak environmental adaptability, and it is difficult to effectively detect the target under the background of high-intensity clutter. In response to the above problems, combined with the powerful learning and representation capabilities of deep learning in image recognition and other fields, a radar target recognition method based on stacked bidirectional long short-term memory network is proposed. The network model constructs a data set with radar Doppler-dimensional echo data, uses bidirectional LSTM to extract the forward and reverse information of radar echo data in the time series, and iteratively trains the neural network parameters through the RMSProp optimization algorithm. Effective recognition of low-altitude and slow-speed small targets such as unmanned aerial vehicle. Experimental results show that the radar target recognition based on stacked bidirectional LSTM is better than the traditional SVM classification algorithm and convolutional neural network classification algorithm.